I'm currently building Polarity, an applied research lab building self-improving models, on leave from engineering at UWaterloo.

I'm passionate about building agents and post-training research. The field moves fast and I find that exciting. My take on the future is that the frontier era is almost over, the moat is drying up, and the models that define what comes next will be open-source and built through RL, not gated behind a handful of labs. I want to work on systems that improve from their own experience instead of waiting on the next release.

GitHub | LinkedIn | X

Currently

Polarity (Present)

Co-Founder & CTO

Post-training models to build collective intelligence. Backed by Afore Capital.

Previously

Shopify (2025)

Engineering

Built and rolled out onboarding agents to millions of merchants.

IntouchCX (2025)

Engineering

Built out enterprise internal workflow systems end to end.

UWaterloo (2025)

Undergraduate Research Assistant under Dr. Ken McKay

Built models and solutions for the Chicago Bureau of Police.

(2024 and before)

A bunch of startups and other roles.

Zenith

Sub-millisecond p95 at a billion rows. Beats Postgres and ClickHouse on agent trace benchmarks. 130+ GitHub stars.

Palkia

RAG context engine for retrieving relevant codebase context and generating structured bug reports. Won Google's AI Innovation Award.

Behavioral Hot Paths (2026)

Author, Polarity Labs

Discrete, cost-aware discovery of recurring agent behaviors for self-specializing agents.

PDF: https://shanebarakat.com/research/hotpaths.pdf

QA-Bench (2026)

Co-author, Polarity Labs

Benchmark for evaluating agent QA across bug detection, test generation, and regression testing on 506 real-world tasks.

PDF: https://shanebarakat.com/research/qa-bench.pdf

Omnigrep (2025)

Co-author, Polarity Labs

Multi-turn agentic code search. 33% relative improvement over Claude Code on CodeSearchEval.

PDF: https://shanebarakat.com/research/omnigrep.pdf